In a recent Genome Biology article, Choe et al. described a control dataset for Affymetrix GeneChips. By spiking RNA at known quantities, the identities of all null and differentially expressed genes are known exactly, as well as the fold change of differential expression. With the wealth of analysis methods available for microarray data, a control dataset would be very useful. Unfortunately, serious errors are evident in the Choe et al. data, disproving their conclusions and implying that the dataset cannot be used to validly evaluate statistical inference methods. We argue that problems in the dataset are at least partially due to a flaw in the experimental design.